How to fight AI slop, according to Hany Farid

Summary

---

Settle In with PBS News

Episode: How to fight AI slop, according to Hany Farid

Date: March 3, 2026
Host: Amna Nawaz
Guest: Hany Farid, Professor at UC Berkeley, Chief Science Officer at Get Real Security

---

Episode Overview

This episode explores the rapid advancements and growing challenges of AI-generated media—deepfakes and generative AI—and their impact on truth, trust, and everyday life. Digital forensics pioneer Hany Farid joins to discuss what “AI slop” means, the dangers and benefits around generative technologies, how the threat landscape has evolved, and what individuals and society can do to defend reality.

---

Key Discussion Points and Insights

1. Hany Farid’s Journey into Digital Forensics

• Origins of Concern ([01:23]–[04:51])
  • Farid began his work in 1999, when digital manipulation was primitive compared to today.
  • Initially, his work focused on authenticating evidence for courts of law, not anticipating the explosion of AI fakes and democratized manipulation.
  • “What started 25 years ago as a bespoke narrow field has really changed dramatically.” — Hany Farid [02:08]
• Shift in Landscape
  • The problem expanded from legal evidence to mass disinformation and state-sponsored campaigns, accelerated by the rise of social media.

2. Speed & Scale of the Problem

• Technological Acceleration ([05:08]–[06:59])
  • Tech upgrades leap from 12–18 months to as little as 12–18 days.
  • Information spreads in minutes. The half-life of a typical social media post: about 90–120 seconds.
• Algorithmic Incentives
  • “The algorithms have learned how to spread the most salacious, outrageous, conspiratorial content, because that's what the billions of people online click on.” — Hany Farid [06:46]
  • Algorithms prefer misinformation because it drives engagement and profit.

3. Deepfakes & Generative AI: Definitions and Mechanisms

• What is a deepfake? ([08:14]–[11:58])
  • Farid questions the term “deepfake,” preferring “generative AI.”
  • Capabilities include text-to-image, text-to-video, talking-head video synthesis, and voice cloning—with minimal input required.
  • “We have gone from needing hours of audio and hundreds of images to 10 to 15 seconds of audio and one image.” — Hany Farid [11:19]
• Threat Vector Expansion
  • Anyone with digital presence is now at risk—not just celebrities or public figures.

4. The Double-Edged Sword: Positive and Negative Uses

• Potential for Good ([12:18]–[14:33])
  • A teenager could use these tools to create blockbuster-level creative works.
  • But real-world harm—non-consensual pornography, fraud, child exploitation, and democracy-threatening disinformation—is outpacing the benefits.
  • “Our job is to start putting in guardrails... Let's not repeat the mistake of the last 25 years with how we allow technology to be weaponized.” — Hany Farid [14:04]
• Guardrails Needed
  • Technological, policy, regulatory, and liability frameworks are needed to balance innovation and harm.

5. Real-World Impacts & Everyday Threats

• Individual Targeting ([15:20]–[17:40])
  • Examples: scam calls using cloned family members’ voices, AI-generated explicit blackmail images, and supercharged phishing schemes.
  • Disinformation can affect how people vote and perceive reality.
• Personal Story
  • Amna Nawaz shares being targeted by an AI-generated video that reached over 17 million views before she could respond.

6. Can You Spot a Fake?

• Detection is Difficult ([18:06]–[20:23])
  • Farid’s research shows people are “not much better than flipping a coin” at distinguishing fakes from real images, audio, or video—even in controlled settings.
  • “Using your senses, you have to understand, is at best limited.” — Hany Farid [18:51]
• Practical Advice
  • Awareness is vital: know that sophisticated fakery happens.
  • Analog solutions: Farid suggests using family “code words” to verify urgent or suspicious communication ([20:23]).
  • Always “hang up and call back” to verify a suspicious phone request.

7. The Lab Perspective: Digital Forensics at Scale

• Defensive Challenges ([21:14]–[23:41])
  • Defenders are always behind attackers.
  • Current focus: using biometrics (face, voice) and contextual cues to authenticate people in images and video.
  • Detection is improving but never foolproof: “We don't eliminate threats, we mitigate them.” — Hany Farid [22:36]

8. AI Misuse: The Hallucination Problem

• False Confidence in AI-Enhanced Media ([23:41]–[28:16])
  • Cases where AI-enhanced images have been mistakenly used to identify people (e.g., masked law enforcement or suspects).
  • AI “hallucinates”: it confidently creates plausible-looking but completely false images or text responses.
  • “This is the mother of all hallucinations. AI doesn't know what somebody under a ski mask looks like.” — Hany Farid [25:15]
  • The public and press often trust AI’s output more than they should.

9. Government & Institutional Failures

• United States’ New Participation in Disinformation ([29:31]–[31:41])
  • Farid notes it is new and alarming to see the US government distributing manipulated political images.
  • “It is wholly inappropriate for the President, United States, and for the administration to be posting videos of the Obamas the way they did... it demeans the White House.” — Hany Farid [29:49]
  • Long-term effect: erosion of trust in institutions.

10. Restoring Trust & Cultural Change

• Trusted Information Sources ([32:04]–[34:30])
  • Farid urges: “Stop, for the love of God, trying to get news and information from social media... It is not a place to become an informed citizen.” — Hany Farid [32:26]
  • Speed comes at the cost of accuracy.
  • Transitioning back to fact-based, slower, professional journalism is essential, but will require a massive cultural shift akin to smoking cessation campaigns.

11. Accountability and Policy Solutions

• Corporate Responsibility ([36:30]–[38:41])
  • Real change happens when penalties outweigh profits.
  • “When did companies start making better products that were safer? ... When the liabilities law said you create a product that you knew or should have known was going to create harm, we are going to sue you back to the dark ages.” — Hany Farid [37:34]
  • Major lawsuits (e.g., against Meta) could create an “awakening.”

12. Generational Perspectives & Policy Tools

• Young People’s Skepticism ([38:41]–[40:47])
  • Today’s young people are more skeptical by default and show analog nostalgia (e.g., Polaroids, flip phones).
  • Policy suggestions: ban social media for under-16s, hold tech giants liable, and break up monopolies to foster safer tech.
• Empowering Parents
  • Legislation can give parents the tools and authority to protect children from harmful platforms.

13. Building Defensive Capacity

• Shortage of Digital Forensics Experts ([43:20]–[45:13])
  • Not enough training or funding for defenders; most energy, talent, and investment goes into generative/AI offensive technologies.
  • Defense is less lucrative but desperately needed; young people’s idealism offers some hope.

14. What Can Individuals Do?

• Personal Agency ([45:52]–[47:51])
  • Learn about AI and digital manipulation; ignorance is not an option.
  • Vote for politicians championing tech responsibility and support responsible companies.
  • Consumer protest works: mass subscriber cancellations can reverse corporate decisions.
  • “We have power even when it doesn't seem like that. And so let's, let's exercise it. Let's demand more of our corporate overloads. Let's demand more of our elected officials.” — Hany Farid [46:37]

15. Staying Motivated in a Tough Battle

• Farid’s Motivation ([47:51]–[49:26])
  • Some days are tough, but giving in to defeatism is not an option.
  • Everyone doing their part is the only viable path forward.

---

Memorable Quotes

• On Tech’s Fast Pace:
  “We used to measure these things in 12 to 18 months... now it's 12 to 18 weeks, sometimes 12 to 18 days.” — Hany Farid [05:30]
• Algorithmic Incentives:
  “The algorithms... prefer algorithmically the spread of mis and disinformation because that's what leads to user engagement.” — Hany Farid [06:41]
• Human Ability to Spot Fakes:
  “The average person is not much better than flipping a coin.” — Hany Farid [18:51]
• On Using Social Media for News:
  “Stop, for the love of God, trying to get news and information from social media.” — Hany Farid [32:26]
• Systemic Solutions:
  “We don't eliminate threats, we mitigate them.” — Hany Farid [22:36]
• Policy Analogy:
  “The parallels to tobacco are not far off.” — Hany Farid [34:59]
• Individual Power:
  “We have power even when it doesn't seem like that.” — Hany Farid [46:37]

---

Key Timestamps

• 01:23—Farid’s origin story and the evolution of the threat
• 05:30—Speed of change in technology and information spread
• 08:14—What is a deepfake and how are they made?
• 11:19—The expanding threat to everyday people
• 14:33—Positive creative potential of generative AI vs. harms
• 17:40—Real-life scams and threats facing ordinary people today
• 18:51—Studies show people can’t really spot fakes
• 20:23—Analog defenses: code words and callbacks
• 22:36—Forensic challenges and the never-ending defense game
• 25:15—AI hallucinations: making up faces and events
• 29:49—US government participation in manipulative media
• 32:26—Advice: Don’t trust news from social media
• 34:59—Parallels to tobacco, need for lawsuits and regulation
• 37:34—Product safety lessons for tech accountability
• 39:23—Generational shifts in skepticism
• 43:41—Need for more digital forensics experts
• 46:37—Grassroots power and consumer activism
• 48:10—Why Farid keeps faith in progress

---

Actionable Takeaways

• For Individuals:

  • Learn basic digital literacy about AI and media authenticity
  • Use code words and callbacks before acting on emotional urgent requests
  • Limit reliance on social media for news—consult trusted sources
  • Voice opinions and vote for leaders prioritizing tech responsibility
  • Consumer action—cancellations, feedback—do make a difference

• For Society:

  • Pursue policy reform: regulation, liability, youth protections, and antitrust
  • Invest in digital forensics education and defenses
  • Hold institutions and officials accountable for transparency and accuracy

---

Tone of the Conversation

Insightful, urgent, occasionally humorous, and ultimately hopeful—with both speakers emphasizing practicality, personal responsibility, and the collective power to drive positive change even in the face of rapid technological upheaval.

---

Transcript

[A]

Hi everyone, it's Amna Nevaz. Welcome to another episode of Settle In. Well, now that anyone in the world who has a smartphone or a laptop can produce a hyper realistic deepfake, how can you trust anything you see or hear online? We're going to get into all of that today with an expert named Hani Farid. He's a professor at UC Berkeley. He focuses on digital forensics and misinformation. He's also the Chief Science Officer at Get Real Security, that's a company that focuses on detecting deep fakes. We talked about how fast the technology has moved just over the last 20 years, how hard it is to spot something fake online and also what people in their everyday lives can be doing to protect themselves and also get better about what's seeing, what's real and what's not online. So settle in and enjoy my conversation with Hani Farid. Hani Farid, welcome to Settle In. Thank you so much for making the time.

[B]

It's great to be with you, Amna.

[A]

So we're going to talk about AI generated images, we're going to talk about digital forensics, about deepfakes. You are, when people talk about you, they attach words like renowned, expert and pioneer. So I'm really excited to make sense of a very confusing world with you. But before we get into all that, just tell me a little bit about you. What led you into this line of work in the first place?

[B]

What's amazing about the work I do today here in 2026 in the world of AI and deep fake and social media is we didn't really see this coming 25 years ago when I started my academic career. So I like to tell students I started my academic career in the previous millennium, which is in fact true. I started my career in 1999. And it's hard to remember 1999, but we largely lived in an analog world. Digital cameras were still on the rise up. Social media was non existent, Internet was burgeoning. And I started to sort of see, see a digital world in our future and started to think about how we can authenticate digital media. Because at least with analog, we had some sense that when we recorded a film or a photo of a voice recording, it was reasonably we could trust that. And so what started 25 years ago as a bespoke narrow field has really changed dramatically through a few waves. So the first wave was going from an analog to a digital world. That's the early and mid 2000s where the Internet became part of our lives. Social media became part of our lives. Mobile became part of our lives. Things like Photoshop and Adobe products for manipulating images and video started to become very sophisticated. But then of course, Starting in around 2015, 16, we saw a whole new wave of this with the power of AI. And suddenly we went from sophisticated state sponsored actors could manipulate reality to anybody with a keyboard and an Internet connection. And suddenly we are now looking at a landscape where anybody can create images and video and audio of anybody doing or saying anything. And so my, my academic career, you know, sort of has followed this trajectory from the analog to the early digital to the now AI. And what we focus on in my lab here at UC Berkeley and previously at Dartmouth College was developing computational and mathematical techniques that can analyze images, audio and video and tell you whether they are real or not and if they've been manipulated or altered in some way.

[A]

Way and way back in the 1900s when you started this work, I mean, when you say you were kind of focusing in, on authenticating this idea of figuring out what is real and what is not, was that you worrying about what was coming down the pike? I mean, why focus on that part of it?

[B]

Yeah, yeah, it's a great question. I started worrying about it because of the courts of law. That was really all I was worried about. I wasn't worried about disinformation, I wasn't worried about state sponsored actor, I wasn't worried about deepfakes. I was thinking about courts of law, which is when we introduce evidence into courts of law, how do we authenticate it? And it was really quite narrow and quite focused. And then what happened over the years and now decades is that we started to see citizen journalism. We started to see reporting around the world, not from people like you, but from people with a mobile device. And then of course we started to see disinformation campaigns and then we started to see state sponsored actors. And so it just kept growing and growing. Our, as we call it, convex halls kept opening and opening and more and more things started to become relevant. But the first days of it were really was courts of law. We were worried about evidence in courts of law. And it was, I long for those days because say what you will about the courts, one of the benefits is that they move slowly. So we had time, we could think, we could analyze, we could create. But now something happens and you know, stopwatch starts and you've got seconds or minutes before these things go supernova viral. And suddenly, you know, our world got very, very big and very, very complex.

[A]

I mean, the speed with which things move now and the speed also with which the technology has developed, you mentioned sort of the democratization of that ability, that technical ability across millions and millions of people too. How would you characterize that? Like, where were we back then versus where we are now?

[B]

So two, two speed dimensions. So one is the speed with which technology is advancing. And we used to measure these things in 12 to 18 months. The cadence of upgrades to the latest laptop, the latest device, the latest version of photoshop was roughly 12 to 18 months. We now measure.

[A]

It's like every new iPhone release, right? Like when new phones are coming out, we're lined up outside stores, right?

[B]

Yeah, exactly. But now it's 12 to 18 weeks, we measure it and sometimes 12 to 18 days, the cadence is so much faster. Now, in fairness, we are in the very steep part of this AI revolution, but I don't think this is going to change in the coming years. So that's speed vector number one. Our, our, our adversary, if you will, is moving at a much, much faster pace. But of course, the other speed is the speed with which information moves online from. We went from the, you know, the 24 hour news cycle to social media and now we measure the half life of a social media post is about 90 to 120 seconds. Half of all views happen in about a minute and a half. And that is a speed that we are not used to without, by the way, any zero editorial control. Right? There's nobody policing these, these networks anymore. And so these things spread like wildfire. And the thing you have to understand about social media is not only is it doesn't care about real fake, true lies, in fact, it actually prefers algorithmically the spread of mis and disinformation because that's what leads to user engagement. So the algorithms have learned how to, how to spread the most salacious, outrageous, conspiratorial content, because that's what the billions of people online click on. And so in many ways we've known this for a long time, is that the lies spread much, much faster than the truth, which of course adds a whole nother complexity to the speed with which we have to respond and also the consequences for getting it wrong.

[A]

I mean, you just said something fascinating though. Not. It's not just that people are spreading these things because they do. Oh, it's not coincidence. Rage bait was the word of the year last year. Right, but it's that algorithms actually prefer them over real information or real images, Is that right?

[B]

That's 100% right. And the reason of course, is because the business model of social media. Think X, Facebook, Instagram, TikTok, et cetera is user engagement. The more you click, the more ads we deliver, the more money they make. And so the algorithms, they didn't set out to burn the place to the ground. They didn't set out to do that. It was learned. And, you know, you could blame the social media giants for this, and I think we should, but at the end of the day, we're the ones clicking on those posts. We are the ones teaching the machines that this is what we will engage with. And so, yeah, it's learned that, well, when the user clicks on this, give them more of this and rage. Bait works. Click bait works. We click on it.

[A]

We're going to talk about all of this in a lot more detail, but let's just back it up for a second for people coming to this without your level of expertise, which will be everyone who's listening, including me. Kind of walk me through definitionally here. What are we talking about when we talk about a deep fake? What are we talking about and what does it take to create one?

[B]

Deepfake is an umbrella term, and I've never actually liked the term very much. Generative AI is maybe the more benign term. So deepfake or generative AI is a broad term for using modern AI machine learning tools to generate images, audio and video. And. And they come in many different flavors. Let's enumerate a few of them. So with images, the one probably people are most familiar with is you type a caption. By the way, I still say please when I prompt AI chatbots. I don't understand it, I think it's weird. But I do one of my please into your prompts always, I swear to God. But does that play the outcome?

[A]

I'm curious.

[B]

I don't think it does. But one of my students told me, and this has haunted me is be nice to the chatbots, because when the AI overlords come, they're going to remember that you were polite to them. And it sort of haunts me when I prompt them. So I'm just going to take her advice right now.

[A]

I'm going to start doing that now.

[B]

Thank you. I know you can't be too careful, Amna. So an image of a penguin wearing a baseball cap walking down Times Square, and it will generate an image exactly of that. So this is what we call text to image generation. And that has been gotten just incredibly good. And there's many, many different AI models for doing that, and you are only bounded by your imagination. Type what you want, and it will generate an image with audio. It goes something like this. I can take about 15 seconds of your voice. I can upload it to a service and say, clone this voice. And then I can type and I can get you in your voice to say anything I want you to say. It's called voice cloning. It's also a type of a deep fake. Now with video, there's many, many different incarnations of this. So one is the same as image to text to image. There is text to video, a video of a penguin wearing a baseball cap walking down the streets of New York. And it will generate a video of that. So think Sora Veo, Sea dance, cling. There's a whole bunch of those. Now there's another type of video, which is what I call a talking head video, where I can now take your voice that I just described, a single image of you. I can take a screenshot right now and then I can drive an avatar of you talking. So I will sit in front of the camera the way I am right now, in real time. It will modulate my voice to sound like you. And as I move, I will animate an avatar of you. And so now I can get on a call, both recorded or in real time, and I can sound and look just like you or a president of the United States or the CEO of a company or whatever I want.

[A]

Wait, wait, wait. I'm going to stop you right there for a second. You need a single image of me and some version of my voice, a few seconds of my voice, and then you could impersonate me in video form. Is that right? Saying whatever you want me to say.

[B]

Yeah, yeah. And the question you gotta be asking yourself right now is, am I recording this? And I mean, I, I could do this right now. In fact, I don't even have. I don't even need this. I just go online. You have enough of a digital presence. Or I could do this with just what you have online. Of yourself. Yeah. And here's.

[A]

How easy is that?

[B]

We were talking about the sort of, the historic trend of deep fake. So it used to be 10 years ago that the people who were vulnerable to this type of attack were, well, people like you who have a big digital footprint or the Scarlet Johansson's of the world with a huge digital footprint. But now we have gone from needing hours of audio and hundreds of images to 10 to 15 seconds of audio and one image, which means that the threat vector has gone from people like you and Scarlett Johansson to, to, well, anybody, Anybody who has any type of social media presence or online presence. And that's a very different threat vector for us as a public who is vulnerable to this.

[A]

I mean, I went immediately to the that's terrifying reaction because it really alarms me in terms of the stories we cover and the kind of things we focus on. I'm just gonna ask us to put it out there. Is there a good case usage here? Like is it mostly fear built around this idea or do you see the potential for good in this technology as well?

[B]

First, I love this question because I think it's an important one and we shouldn't be fear mongering. So a couple of things. One is when I describe non consensual intimate imagery, child sexual abuse, small scale fraud, large scale fraud, disinformation, these are not hypothetical, these are happening every day to individuals, organizations, societies and democracies. So that threat is not, oh, this may happen to us at some point in the future, that this is happening today and it will continue to happen, number one. Number two is to your question now, are there interesting and positive applications of the technology? Sure, of course they are. It's incredibly creative. I can imagine, by the way, in the coming years that the next blockbuster movie will not come out of Hollywood, but it will come out of the mind of a highly talented 15 year old in their bedroom working with generative AI who can now create entire scripts and videos and music and sound. And it's going to be amazing. There are really, really exciting applications. But now the question you want to ask is the counterbalance? Okay, let's factor in the creative side of things and then the threats to individuals, organizations, societies and democracies. And I would say the threat is right now outweighing the social benefits. But that's okay, because now our job is to start putting in guardrails. Both technological, both policy, both regulatory and both liability to counterbalance. Everything in our world has positive use cases and bad use cases. And our job is to figure out how do we counterbalance. And every once in a while we say, okay, look, the bad use cases are simply higher than the good use cases. We're not going to allow this technology until we can figure out how to counterbalance. So I, and nobody is saying get rid of generative AI, get rid of AI. But what we are saying is let's not repeat the mistake of the last 25 years with how we allow technology to be weaponized. Let's put in thoughtful guardrails, by the way we do in the physical world when we release products, when we Release medicines and we sell food and we build airplanes and we sell products, electronics. There are guidelines, there are safety guidelines that say yes, this stuff is great, but if we're not careful, it's going to kill us. Okay, let's do the same thing with the Internet. The Internet has had its time of move fast and break things. We now recognize that that policy simply doesn't work.

[A]

I want to talk a little bit more about the guardrails and what you see as solutions, bright spots ahead, but also to make it real for people because sometimes we do focus on the very high profile instances in which we see deepfakes, generative AI having some sort of negative impact. Right. I mean I haven't shared this story with you before, but I was the target of an AI image and video online because someone took my voice and made it sound like I was saying something I was not and put it over at fake video. Now thankfully the video was so fake you knew it wasn't real. It was like recreating news events that didn't actually happen. But it was my voice and it was clearly enough my voice that people flagged me because I have a recognizable voice. And it was, by the time it was flagged to me, viewed more than 17 million times.

[B]

Totally.

[A]

And it was incredible to try to see efforts to try and take it down. But in terms of what people may see in their everyday lives, not high profile people or politicians or anything, what, where, where can they be targeted? When you talk about the threat vector for everyday Americans, where is that?

[B]

Yeah, I'm glad you asked that. And look, maybe, maybe we accept as a public figure there's some amount of risk to that for people like you. I, I, I don't necessarily think that's an okay thing, but okay, I'm, I'm going to concede that point for a moment. So here's the things that we are seeing for people not like you, the average citizen. So we are seeing things like parents and grandparents getting phone calls in their kids voices and saying, mom, dad, grandma, grandpa, I've been hurt, you need to send $2,000 or you know, I won't be able to get my car out of the impound or I'm going to jail. And we have seen these types of scams, particularly hitting senior citizens who are already vulnerable to these types of cyber attacks. That's number one. We've seen children, horrific things happening to children where people will take a photo of a child that is posted online, create, make up an explicit version of that where they are for Example, clothing has been removed. It sends it to that kid and said, if you don't do this, we will do this. And horrific things are happening to those, those children. So the small scale fraud to adults, horrific things that are happening to children is happening today. Think also, all the phishing and scams that you've been seeing on your text messages and on your email are now being supercharged by generative AI where you're not getting a text message from what seems to be your boss, but you're getting a voicemail or a video call from them. And it is not who it purports to be. So a lot of the cyber threats that we have been dealing with over the last 10, 20, 30 years are now being amplified and accelerated by both AI bots that are driving them. So you don't need a human in the loop. And also AI powered in terms of the deep fake and those, these are not, by the way, hypothetical. These are happening. And the last one I will say, particularly since we have a midterm election coming up here in the United States, is disinformation around what's happening in the world. And that very much affects all of us as individuals. It's how we think, how we reason, how we vote and what we know about the world. And that is absolutely impacting every citizen, particularly those who get the majority of their news on social media.

[A]

So you run a expert led sophisticated lab and team to work on the digital forensics piece of this, the authentication piece of this for everyday Americans, if someone gets a call from someone who sounds like their grandchild crying and asking for help in $2,000, or they see an image of someone who seems to be someone in their family in a compromised position, how can everyday people understand what's real and what's not?

[B]

So there's bad news and more bad news. So the first bad news is in addition to developing computational mathematical techniques to detect these things, we also run perceptual studies. We show images and audio and video to the average person on the Internet and we ask them, can you tell the real from the fake? And this is, we are setting you up for success. We're putting you in a lab environment, we are telling you exactly what to do. We're telling you that some of the things you're seeing are real, some of them are fake. And we simply try to see this versus a call at 2 in the morning or rage scrolling on social media where you are not set up for success. And what I can tell you is that when it comes to Images, audio and video. The average person is not much better than flipping a coin. They're basically at chance it's over. And even if it wasn't over, and even if I could tell you, look for this, listen for this, that advice will simply not work three months from now when the next generation of technologies emerge. And so using your senses, you have to understand is at best limited. Okay, so that's the bad news on top of the bad news. The slightly better news is this conversation right here. Raising awareness, letting people know that when they get a phone call from what sounds like their loved one, it may not be them. Knowing that when you get a video call or what you see online may not be real, I think is really important. So public awareness. Now I will tell you what my wife and I do is we have a code word. So if one of us gets a call from the other saying something that is out of the ordinary, we have a code word. And it's only the two of us. I have actually code words with other people I work with in sensitive cases. And I love this because it's an analog solution to a very modern high tech problem. And so you can have a code with your family. In fact, you should do this. And it's super easy to do. You can also hang up. I know that sounds harsh, but call back. Because a lot of time what is happening is that the phone numbers are being spoofed, and when you call back, it is not what it purports to be. So awareness that this can be going on is really, really important. And by the way, that's true with spam and phishing and malware and ransomware. A lot of this is don't click on a link from an email that you've. Not that you don't know. It's a lot of this is just we have to keep talking about this as a psa, but then starting to put some guardrails in place. You know, code words are super, super easy to do. Callback is super easy to do.

[A]

Code words with your family. So, meaning if you don't hear this word, you know, this is not someone that you trust. Is that right?

[B]

Yeah, yeah, exactly. Yeah. Super easy. Yeah. Doesn't take anything. You got to remember it. I have to. I check with my wife every few weeks. I'm like, what's the code word?

[A]

Well, it's almost like in the movie Inception, like the totem, right? It's like the one thing you can turn to to figure out if you're in the dream or if you're in reality something you come to again and again.

[B]

I don't love the analogy, but yes, because it's. Yes.

[A]

Try to make it relatable for the rest of us here. Hani well, let me ask you this, because the fact that it's evolving so quickly, the technology is developing at such a speed, you tell us a little bit about what we can do as everyday citizens trying to figure all this out, but you guys are working at the highest levels here at your lab. Tell me about what your lab is doing. How are you responding to this in real time?

[B]

Yeah, this has always been a hard problem. Playing defense is always harder than playing offense because the way this works is my adversary, if you will. They get to work in quiet, they get to work for as long as they want, and then they release a product and then my starter gun goes off and now we have to start playing catch up. So we are fundamentally always behind, always. That's just the nature of what it is. So what we have been focusing on is we want to be able to detect fake images, fake audio, fake video, to do it both in real time and be able to do it for the types of things that we get that we find online. But what we have decided to do, both in my lab, but also I co founded a company called Get Real Security. So over at Get Real as well, is to focus on images, audio and video of people. We're just going to narrow the scope a little bit for now because we are so overwhelmed by the complexity of the problem. So when we get a video of a person behind a microphone talking, how do we authenticate her voice? How do we authenticate her video? And if that person says she is Omna, how can I make sure it is so can I use biometric facial identification? Can I use biometric voice identification? Are there things about your environment, for example, that I know about, that I've seen you in this environment before? The things about that microphone I know about and, and using that type of knowledge about an individual to protect the Omnis of the world, the CEOs of the world, the members of the Senate, Foreign Relations of the world. And so the, the good news and the bad news is, so the good news is we are getting better at this and the defenses are getting better. The bad news is, and I think this is not going to shock anybody. They're not perfect. Nothing is 100% right. And so I think it's really, really important in this space to, to not think of cybersecurity as an elimination. We don't eliminate threats, we mitigate them. And that's true in the physical world, too. I lock my front door when I leave the house in the morning, but I'm also aware that somebody can still break into my house, but I've made it more difficult. And I think that's the way we have to think about this, is that this is one layer technology. That's what we do. And then, of course, there is education. That's what you and I are doing right now. And then I think we need our government to wake up and start thinking about a regulatory landscape to put some guardrails on this technology. I think we need the corporate world to start taking more responsibility for how their technologies are weaponized. And the goal of those four things is to come together and to take advantage of this amazing technology, but also to keep us safe.

[A]

One of the places people may have seen this in headlines and in news recently was, and something you've been raising the alarm about was people using AI enhancement tools to try to identify people in videos. Right? We saw this in Minneapolis with people trying to use these tools to identify the masked ICE agents who had shot US Citizens there. We saw it with people trying to enhance video and identify the man who shot and killed Charlie Kirk. You've said that this is a wholly inappropriate use of the technology. Tell me about that. Why?

[B]

So I have to say I've been doing this for a long time, and I'm pretty good at it. And I didn't see this coming. I didn't see this application of AI and how it was going to be misused. So. So let me, let me set the stage a little bit, and I'm going to talk specifically about this. So setting the stage. We are seeing three types of applications of AI. People using AI to whole cloth create events in people who don't exist. So just outright lies. This is not who it purports to be. This is not a shark swimming down Route 1 after a thunderstorm. We are also seeing AI being used to recreate events that actually happen. So, for example, when the US Invaded Venezuela and took the President Madura, people were creating fake images and video of that event. The event happened, but what we were sharing were not. And that one's a little bit more complicated, right? Is it. Is it false? No, that did happen, but that is not an example of or an image of that happening. And then number three is, as you were just describing, is that now people on the Internet have taken to using AI to, quote, unquote, enhance. Think the. The the, the, the, the. The most silly stereotype of your favorite crime drama on television. You hit a button and you go from 4 pixels to an HD image, right? So what we have seen people do, for example, is take an image of an ICE agent who killed Renee Good. And he was wearing a mask, so his face is covered from the, the base of the nose down and said, we want to identify him. Ask AI to do that. AI doesn't know anything. So it simply does it and it creates an image that looks like a person. It just wasn't the ICE person. In fact, it led to a misidentification and that person started getting death threats. Same thing happened with the Guthrie case. In fact, arguably worse because that was an image where somebody was wearing a full blown ski mask and they gave that image to an AI and said, what does the person look like underneath? The AI gladly did that and the person responded, well, this is what I got, better than nothing. And, and my response was, no, this is worse than nothing because you are absolutely making stuff up. This is the mother of all hallucinations. AI doesn't know what somebody under a ski mask looks like. This isn't magic, right? This isn't. I don't, I don't even. This is. I don't think the people doing this are trying to create harm. I think they simply don't understand the technology. And because the technology is so enticing, you ask it a question and it's very certain of an answer, even when it doesn't know the answer. And so you can see how this technology is being misused. And we've seen multiple cases of this.

[A]

Can I stop you for a second there? You used a word there that I want to make sure people understand. Hallucinations. What are you talking about? When you mention hallucinations in AI?

[B]

Yeah, so this is a term. I don't actually love the term, but it is the term that we use where you ask AI something like your favorite chatbot, like ChatGPT or Gemini or Claude, and you say, tell me about X and it says something very definitively, but it is made it up. It's actually not true, simply untrue. And that's why the term hallucination is, is overly generous. It's just wrong. It's just wrong. Now, when it comes to images, it will gladly take an image of a person that is masked and generate an image of them unmasked. I can tell you we've done the study where when you do that, it is completely wrong. It has no relationship to the Biometric identity of the person it started with. And, but you know, when you look at the image, it's enticing it. Look, it's accurate. I mean it's not accurate. Sorry. It's high resolution and it looks like a person. And so people are like, oh, this is, must be what they look like. And I think that, I don't think it comes from a bad place. I think it comes from just not understanding the technology and certainly not understand the limits of the technology. And I think this is a sort of a whole new level of danger that we are seeing. And we saw it with Charlie Kirk. Kirk. We saw it with Minnesota, we saw it with the Guthrie abduction. And we're going to see more of this. There's another one where somebody, it was coming out of the Epstein files where somebody took a profile photo of somebody like this and said, what do they look like? Straight on. No, no, no, no, no, no, you can't do this. This doesn't work. This isn't csi. Push a button, enhance. It doesn't work this way.

[A]

So it's people trusting the AI more than they should because they asked it a question, it gave it an answer and they think, well, this is the technology, so it must be right.

[B]

Yeah, I think that that's exactly the way, the right way to put it. They trust it. Right? And they don't. And they trust it because they don't understand it.

[A]

There's also we've seen instances of very intentional image manipulation. We see this in political circles often as well. We've seen even at, you know, the level of this administration, this government using AI generated using digitally altered images and videos on their social media platforms. Some of them are very obvious. Right. Like I know recently we saw the very racist video featuring the Obamas and a bunch of other people as animals and characters from the Lion King. But others have been photos that have been doctored and been altered. And it takes time for the news and reporting around it to show that it was altered. And I'm thinking about the woman, Nakima Levy Armstrong, who's crying when she was detained. Her photo in which she was not crying was altered to make it look like she was crying. This is one of the women who went in as a protester to the Minneapolis church. Is that level of intention and dissemination at that level? Is that something we've seen before?

[B]

Well, we have certainly seen state sponsored disinformation campaign from Russia, from China, from North Korea, from Iran and from around the world. We have not seen it from the United States So in that regard, it is new. It is new to see this level of deceit coming out of the White House. And look, I think it is wholly inappropriate for the President, United States, and for the administration to be posting videos of the Obamas the way they did, or videos of Donald Trump in a jet plane dumping brown substance onto protesters. Wholly inappropriate. And I think it demeans the White White House, it demeans the office, and it means, why should I ever trust anything that you post anymore? And there's going to be good reason for us not to.

[A]

Well, walk me further down that for a moment, because when you put it that way, it doesn't make sense, Right? If there's a body that wants to be trusted, that wants to deliver a message to people, why continue to undermine it by putting out images and videos that call into question whether they're real or not? What, when you've seen that strategy in other places, what has it led to? Why? Why is that done?

[B]

Yeah, I don't have a great answer for you. I think the generous interpretation is they're not thinking ahead that short term, we think this is funny, there's a political gain to be had, and we simply aren't thinking forward. I don't have a great answer for you. And honestly, when the White House was asked, why did you do this? Their response was disheartening. The response was, well, why not? So I just don't think they're thinking this through very carefully. And that is. That is worrisome. And it's not just worrisome. Look, this administration will come and go. We know this. But you are eroding trust in our institutions that we need not just for the next three years, but potentially a long time out. And that is. That is really worrisome. Look, when one mainstream media outlet gets it wrong, it creates a shadow over all of them. Right? We're all in this together. We can disagree on lots of things. It's okay to have political disagreements, but we have to be able to trust institutions and organizations, and we all have a huge, huge reason to want to do that, regardless of our political affiliation.

[A]

We have seen you mentioned the spread of misinformation and disinformation and how much that's changed just over the course of your academic career. And a lot of that was before generative AI became as good as it is and as widespread and as easily accessible as it is. I think anyone listening right now is just going to wonder, like, how can you trust anything you see anymore?

[B]

So, first of all, 100%. Right. Look, even in the analog world we had mis and disinformation, but it was reasonably controlled, right, because we didn't have the Internet, we didn't have digital, we didn't have deep fakes in AI. And then of course, as the digital revolution came, that started getting amplified. Social media amplified it more and deepfakes and generative AI amplified it even more. And so everything is a multiplicative amplification factor. And here we are where I'm going to get to your question. How do you believe and trust anything you see anymore? And I think the short answer is on social media you don't stop, for the love of God, trying to get news and information from social media. It's not what it was designed for. It's not what it is good at. It's, it is not a place to become an informed citizen. If you want to be entertained there, if you want to be rage baited there, I think that's a problem. But fine. But that is not a place to figure out what is going on in the world. I mean, how many years do you have to get lied to before you start getting mad and start saying, okay, I give up. Yeah. So I think, and look, I'm not saying media outlets are perfect, but they get it right way, way, way more often than they get it wrong. And you can't say that about social media. You can't even say that about the CEO of some of the social media companies. They get it wrong way more than they get it right. And so we have to return to our trusted sources. We have to understand that people like you work really hard to figure out what is going on in the world. They talk to people like me to understand it and to bring that information to you. And you don't have to figure out what's going on in the world in the first 10 seconds of an event. There is a speed, accuracy, trade off. The faster you get your information, the less accurate it is. I would much, much rather wait till tomorrow morning and know what is going on than be lied to for the intervening 12 hours. And we just have to change that culture. That people have to understand that social media is not designed as a reliable source of information. It's not. It never has and it never will.

[A]

Can I ask to follow up on that though? And thank you for the very kind things you say about me and my colleagues here. But the majority of Americans do get their news and information from social media. That's where we are right now. What you're calling for is an Enormous cultural shift, the likes of which we are definitely not trending towards. Right. So just culturally, short of even policy guidelines or companies and CEOs completely changing how they do their work, how does that kind of thing start?

[B]

Yeah. So first, I'm not naive about. This is a massive cultural. And not just here in the United States, but globally. And arguably what we are seeing outside the US Is even. Is even worse in some ways because of the echo chamber of how social media has dominated other parts of the world. I am aware of that, but I'm also old enough to remember when something like one third of Americans smoked cigarettes and we thought, well, that's just the way it is. And we smoked in grocery stores, we smoked in restaurants, we smoked in airplanes, for God's sakes. And it was the norm. But we changed because we realized that this stuff is killing us. And I would argue social media is not that different now. I don't think that change will happen slowly. I don't think it will. It will come without pressure from Congress, from our regulators, with liability from the courts holding these companies responsible for the harms. The parallels to tobacco are not far off. We have had things in our society that have been decidedly bad for us for a very long time, and it took decades pushing against corporate interests, understanding what is happening here. And I think this will take that amount of time. Now, the good news is I think there's some, at least a glimmer of hope in. In the horizon. So if you look, for example, today, there are massive litigations happening around social media and the addictive properties and the impact to children in a way that I think 10 years ago I didn't think we would see these cases. And so there is movement. Australia has banned social media for kids under the age of 16. The EU and the UK and other parts of the world are considering similar legislation. I think there is an awakening that while there are positive aspects to these technologies, to social media, it is clear the harms are unambiguous. And now we have to start responding. I think we're late to the game, but at least we're moving in the positive direction. So I think there is some positive movement here, but it will take a lot of conversations, it will take a lot of serious people thinking about this in a serious way. And it will take fighting back against massive, massive global corporate interests. But I don't know what the other option is.

[A]

You mentioned that some of the court cases right now, we've been covering those, and the landmark one right now really seems to be the one in which CEO Mark Zuckerberg, Meta just testified recently and you know about the socially addictive nature of Instagram and other platforms as well. And we've seen, seen other social media companies settle. Right. They said they don't want to go to court. Facebook and Meta seem to be fighting this. They're facing thousands of other related lawsuits. Do you feel like that and this moment could lead to a change in the incentive structure for CEOs for the people who've been driving this content?

[B]

Yeah, and I do. And I'll tell you why. Because up until fairly recently, these social media companies have been able to do whatever they want with very, very little financial implications. Yeah, they pay a fine every once in a while, but it is nothing. It is nothing compared to profits. When you effect change is when corporations have to internalize. Not million dollar fines, not $10 million fines, but $10,000,100 billion fines. When they start to internalize that type of financial penalty is when the business model changes. But up until then they're doing a calculation. Yeah, we get PR, bad PR. We got PR people for PR problems. Yeah, we pay a $10 million fine. We don't care. It's a rounding error. It's what we pay on paperclips every day. Doesn't matter. But when this starts to become 10% of global profits, then the calculus changes. They internalize that liability and they change the business model. We know this, we know this from product safety. When did companies start making better products that were safer? Is when the liabilities law said you create a product that you knew or should have known was going to create harm. We are going to sue you back to the dark ages. And products got safer, medicines got safer, food got safer, transportation got safer. We're going to have to do the same thing with social media. We let them off the hook for 25 years. We've created these monstrosity, global monstrosities. But I think there is an awakening. And I think it started outside the U.S. but now, thanks to some of these lawsuits and other similar ones, I, I think there is going to be an awakening. And it's. Look, it's not a complicated calculation, right? The fines have to outweigh the profits. That's it. That's it. And then change happens. It's not complicated.

[A]

Is there something generationally too that gives you hope? And I ask this because. So I'm in my mid-40s, I'm part of the generation that grew up in my childhood analog and that lived most of my adult life in the digital world. Right. So my skepticism has been learned over time. Time. My children are very skeptical of what they see sometimes online. And I mention this because literally my 10 year old the other day saw a photo, an actual photo of a friend's wedding. And it was a beautiful photo. It was so beautiful and perfect that she looked at it and said, oh, that's AI. That was her first reaction. And so I wonder, for the generation that's growing up fully digital, fully aware, fully familiar, is there a better sense of skepticism?

[B]

Yeah, yeah. Two things. One is yes, but we have to be careful too. We don't want to get over skeptical. We don't want to say everything is fake. Right. So we want to be careful. But here's what I can tell you. I'm on a university campus, and one of the joys of being on a university campus is you get to see every generation as they are coming through and there is a new generation coming and they are thinking about this differently. I also think they don't, they don't care about social media. They're not on Facebook. There's a little bit of Instagram, a little bit of TikTok. But you know, they're also, I see kids with Polaroid cameras, analog cameras, doing this with their photos. It makes me.

[A]

That's amazing.

[B]

Flip phones are back right there. I think there is a realization that social media is toxic. And I think this generation is starting to use technology in a very different way than the previous technology, the previous generation and the previous generation. So I do think there's some hope here. I think we have to help them. I think we have to enable them. One of the great things about the Australian ban on social media is that an empowered parents to say, hey, look, I, you know, because now you know what happens. All your friends are on social media. Okay, I got your. Now you are on social media. But parents have cover now. And this is what the governments have to do. They have to empower parents. You can't say, you can't let Mark Zuckerberg getting away with saying, oh, we've got parental controls. You can, you can do what you need to. No, no, no. This has to come from the top down. So I am hopeful that this generation, I think is being more critical, is being more thoughtful, has learned from the lesson of the previous generation. But I think we have to empower and enable them.

[A]

So social media bans are definitely another tool. When you look at other potential policy steps that could be put into place, what else can leaders do right now?

[B]

So I Like the ban for 16 under, I think it should be 30. But I'll concede 16 is a fine age for now. I mean, by the way, this isn't an outrageous thing. I mean we all kinds of things we don't allow kids to do before, before they're of an adult age. So this doesn't seem particularly outrageous. And the studies are overwhelming evidence. The former Surgeon General has said the time for debate is over overwhelming evidence that social media is net negative for kids. Are there positive aspects to it for certain marginalized groups within that age group? Of course there are, but you have to look at the broader issue here. So in addition to putting bans or restrictions on 16, 18, whatever you want that age to be, I do think there are things that can do. So one is these tech companies have to stop being let off the hook. From a regulatory perspective. We have deregulated the tech industry in a way that we would never allow in the physical space. And we need to start thinking about how these tech companies should be held responsible for creating safer products liability. I think when the liability starts coming down hard from the courts, I think it's actually going to start to affect change. And then here's the big one, the winners from social media, the last, you know, the five giants of tech are going to be the five winners in AI because those are inextricably linked. We have virtual monopolies and we need to allow for competition in this space. You can't allow these global monopolies to keep dominating. We have to make room for better business models, better CEOs, better technology, safer technology. And that can't be done with when you have these massive 800 pound gorillas that can squash any competition at will. So I think economically we have to create a lively ecosystem that allows small companies to compete with these massive giants. And by the way, when Google tells you, no, no, no, you can't do that, that is bad for competition. You should remind them that when they were a burgeoning startup, they went to the Department of Justice crying like a bunch of babies because Microsoft, the then 800 pound gorilla, was beating them up. And the DOJ stepped in and it worked. We created Google. You don't get to then close the door and say, well, we don't want that kind of regulation anymore. That's why you exist is because the government stepped in and said, stop using your monopoly to squash competition. And it worked. So let's do it again. Let's protect the companies now from Google, which was protected from Microsoft.

[A]

What about more People like you and the people who work with you and the students that you teach. I mean, I'm old enough to remember when everyone thought you can future proof your career by learning to code. And now there's a program that can do that for you in seconds online. Are we doing enough to train the next generation of digital forensic experts, people like you?

[B]

Yeah, yeah. No. And there's a couple of reasons for that. So one is there has been over the last year a general assault on, on the academy, international students, researchers like me who research deepfakes and disinformation. There has been an assault on our funding, on the type of work we do because it's not aligned with this administration. And that's been a real problem. And that is a decades long problem because you have to feed the pipe, right? It takes a long time to educate these kids. But the other reason is that if you look both at everything from where the venture capital money is to where the young people want to go, it's on the generative side. The young people and the money and the power and the fame is all on the generative side. We are 100 to 1000 to 1 outgunned. The vast majority of students want to be on the other side of the aisle. And part of that I think is financial. They see a windfall there and they don't necessarily see a windfall in the type of work I do. Nobody wants to pay billions of dollars for defense. They'll pay billions of dollars for the offensive tools, but not for the defensive. Now the good news though is since we were talking about young people is I do think that this generation coming up is more idealistic. I don't think they all want to go work at Facebook and Google and sell ads to get people to buy things they don't need. I think they want to have a sense that the work they are doing is meaningful, important. And so I'm hopeful for the generation that is making its way through college now, as opposed to the previous generations that I think got bamboozled by the tech industry.

[A]

You've been doing this work for decades, right? I mean, you can speak about it with such familiarity and some fear, but a lot of some optimism as well in terms of what needs to happen. For anyone listening to this conversation who hasn't really thought deeply about it before, who's trying to find their footing right now, and maybe a little worried because of everything they just heard, heard, what should they take away from this? I mean, what are. You mentioned the code word earlier in terms of Being able to figure out what's real and what's not if you're specifically targeted. But what else should people be doing right now on an individual level so they don't just feel like this is all happening to them in the moment?

[B]

Yeah. Yeah. Well, I'm stockpiling food and water, so I don't know. No, that's not. Sorry, that's. I'm not helping. Sorry. That's a terrible answer.

[A]

Not helpful, honey.

[B]

Okay, a couple of things. One is you got to get smart fast. You can't say, wow, this whole AI thing, this is like saying, oh, the whole Internet thing, the whole technology thing, We've got to get smart fast.

[A]

We've got to get it right. Like it will impact you.

[B]

Yes, this is coming. This is coming. I don't know if it's going to be today or tomorrow or next year, but our lives, both personal and professional lives and certainly the lives of our children and grandchildren are going to be impacted. And I know it is unfair to say, well, you got to get smart about this stuff, but you do. You got to start understanding this. We have got to start voting for politicians who are want to protect us and not protect five people in Silicon Valley. We've got to start voting with our feet and start telling people, look, this is not okay. Right? If you don't like the way Mark Zuckerberg and Elon Musk and Jeff Bezos do things, stop using their products. And I know, I know this seems dumb. You're like, I'm one of a billion customers. What does it matter? But it does matter. The same way voting matters. The same way it matters that you recycle. Small actions matter, right? And start voting with your feet. Start voting with your wallet. Start telling these companies, this is not okay. Let me give you an example of that. When ABC canceled Jimmy Kimmel's show after pressure from the administration, hundreds of thousands of people overnight canceled their Disney and Hulu subscriptions. And it affect a change, they reverse course and two days later he was back on the air. We have power even when it doesn't seem like that. And so let's, let's exercise it. Let's demand more of our corporate overloads. Let's demand more of our elected officials. And I know it seems like while I'm just one voice, but yes, you're one voice, but talk to two people and they'll talk to two people and they'll talk to two people. And that's how change happens. Grassroots can happen, but we've got to Decide we are not going to be held victim to five multi, multi billionaires in Silicon Valley.

[A]

I mean, you are in the thick of it. Every single day. You see the worst of what's possible. You can see ahead in a way that many of us cannot in terms of what more could be coming and you know which way this is trending. Just on a personal level for you, what keeps you going every day? How do you stay in this?

[B]

There are days I'm not. There are days where you're like, I'm in a fetal position under the bed. So look, I, we, I think we're at a fork in the road and it's coming up fast and we're gonna have to make some decisions. And one leads to some weird dystopian future that I don't think is going to be good for anybody except for maybe five people in the world. And the other leads to something that I think can be exciting and powerful, powerful and enabling for society. I think we're leaning in the wrong direction right now. I think we're going down that. It seems like we're leaning in the direction of the path that I don't want to go down. But here's my thing, is what are the options, right? I mean, somebody's got to do something and I'm going to do my little part and I'm hoping that other people are going to do their little part. You're doing your part and I think, you know, otherwise we give up, right? We say up, this problem is too big. But I don't think that's an option. And I think that's true of the things that we're talking about now. I think it's true of global climate change. I think it's true of many of these massive, massive problems. We have got to start, stop being defeatist about this and say we're going to do something about it in our own small way. Otherwise it's the fetal position under the bed for the next few years and that's not an option.

[A]

Well, I find great hope in the work that you do and the way you talk about it and your colleagues as well, find great Hope in my 10 year old skepticism as well. As do I. Hopefully we can turn this around. Hanifarid, such a pleasure to speak with you. I learned so much. I'm sure our listeners did too. Thank you very much for making the time.

[B]

Thanks, Amna. It's wonderful to talk to you. Sa.